Open-source unattended agent framework on GitHub Actions with x402, MCP, and A2A integrations.
- What it does
- A fully autonomous agent framework ("no approval loops, no babysitting") running on GitHub Actions with a self-healing skill loop, x402 pay-per-call settling USDC on Base, base-mcp, and Google A2A support.
- Target user
- Technical solo builders who want a free always-on agent.
- Product loop
- Cron heartbeat → skill execution → scored output → skill repair → repeat.
- Token / revenue model
- No native token documented in the repository; a community $AEON launched via Bankr exists outside it. No revenue model.
- Distribution
- Creator's X following and GitHub.
- Verifiable strengths
- Actively maintained, genuinely open-source, with a catalog of 196 defined skills whose category sums check out.
- Evidence gaps & weaknesses
- All capability claims are self-description; the "40+ ecosystem projects" figure is not substantiated by the README; unattended autonomy with auto-installed skills and auto-payments is precisely the pattern implicated in 2026's agent security incidents.
- Whitespace it reveals
- The trust layer unattended agents lack: skill auditing, action attestation, spend governance.
- Technical primitives
- x402MCPA2AGitHub Actions
“195 skills live across 8 categories.”
The official README (actively pushed as of June 9, 2026) lists 196 skills across 8 categories with a named catalog whose per-category sums check out. This is self-description from the project's own repository: the count of defined skills is supported; whether each skill functions as advertised has not been independently tested.
“Powering 40+ ecosystem projects.”
The main README lists approximately 16–18 named community skill packs and defers any broader count to a separate file. No independent enumeration supporting 40+ was found. In Basis's adversarial review, the claim that the README substantiates this figure was itself refuted.
“x402 pay-per-call settling in USDC on Base, MCP servers including base-mcp, and A2A support.”
All three integrations are documented verbatim in the README of an actively maintained repository. Documentation-level support is confirmed; settlement volume or live usage has not been independently measured.
“Self-healing across 8 categories (scored outputs, heartbeat → skill-health → skill-repair).”
The self-healing loop is described in the project's own documentation. No independent reliability testing, uptime data, or third-party reproduction was found.
“Aeon's README documents no native token; a community $AEON token launched via Bankr exists outside the repository.”
The README contains no token documentation. A Bankr-launched $AEON associated with the creator's account exists on Base. Recorded as context so that 'no token' is not over-inferred from the repository alone.
Represent Aeon? If any label here is wrong, we want to know. Send evidence to corrections@basis.watch. Substantiated corrections are applied, versioned, and logged publicly — see the correction process.